Update llama-instruct-inference.yaml

recipes/wmt22_with_llama3.1/llama-instruct-inference.yaml

@@ -1,18 +1,3 @@
-transforms: [onmt_tokenize]
-
-transforms_configs:
- onmt_tokenize:
- src_subword_type: bpe
- src_subword_model: "${EOLE_MODEL_DIR}/llama3.1-70b-instruct/bpe.model"
- src_onmttok_kwargs: {"mode": "space", "spacer_annotate": True, "preserve_placeholders": True}
- tgt_subword_type: bpe
- tgt_subword_model: "${EOLE_MODEL_DIR}/llama3.1-70b-instruct/bpe.model"
- tgt_onmttok_kwargs: {"mode": "space", "spacer_annotate": True, "preserve_placeholders": True}
- gpt2_pretok: true
- mapped_tokens: [['<|start_header_id|>', '｟start_header_id｠'], ['<|end_header_id|>', '｟end_header_id｠'], ['<|eot_id|>', '｟eot_id｠']]
-
-optional_eos: ['<|eot_id|>']
-
 # Model info
 model_path: "${EOLE_MODEL_DIR}/llama3.1-70b-instruct"
 
@@ -30,9 +15,9 @@ parallel_mode: "tensor_parallel"
 quant_layers: ['gate_up_proj', 'down_proj', 'up_proj', 'linear_values', 'linear_query', 'linear_keys', 'final_linear']
 quant_type: "bnb_NF4"
 compute_dtype: fp16
-#random_sampling_topk: 1
-#random_sampling_topp: 0.0
-#random_sampling_temp: 0.9
+top_k: 0
+top_p: 0.0
+#temperature: 0.9
 beam_size: 1
 n_best: 1
 report_time: true